Let $\left(\frac{a}{p}\right)$ denote the Legendre symbol and $p \geq 3$ a prime number and $A$ an integer with $p \nmid A$. Prove that $$\sum_{X=0}^{p-1} \left(\dfrac{X^{2}+A}{p}\right)=-1 .$$

I wasn't sure how to manipulate this sum. Maybe if we show that $\left(\dfrac{X^{2}+A}{p}\right)$ can't be a quadratic residue for more than some amount of values it will help.


Fix an odd prime natural number $p$. Let $$S_p(A):=\sum_{x\in\mathbb{F}_p}\,\left(\frac{x^2+A}{p}\right)$$ for all $A\in\mathbb{F}_p^\times$, where $\mathbb{F}_p^\times$ is the group $\mathbb{F}_p\setminus\{0\}$ of units modulo $p$. Note that $$ \begin{align} S_p(A) &=\left(\frac{A}{p}\right)+\sum_{x\in\mathbb{F}_p^\times}\,\left(\frac{x^2+A}{p}\right)=\left(\frac{A}{p}\right)+\sum_{x\in\mathbb{F}_p^{\times}}\,\left(\frac{1+Ax^2}{p}\right)\left(\frac{x^{-2}}{p}\right) \\ &=\left(\frac{A}{p}\right)+\sum_{x\in\mathbb{F}_p^{\times}}\,\left(\frac{1+Ax^2}{p}\right)=\left(\frac{A}{p}\right)\,\left(1+\sum_{x\in\mathbb{F}_p^{\times}}\,\left(\frac{x^2+A^{-1}}{p}\right)\right) \\ &=\left(\frac{A}{p}\right)\left(1-\left(\frac{A}{p}\right)+\sum_{x\in\mathbb{F}_p}\,\left(\frac{x^2+A^{-1}}{p}\right)\right)=\left(\frac{A}{p}\right)-1+\left(\frac{A}{p}\right)\,S_p\left(A^{-1}\right)\,. \end{align}$$ If $\left(\frac{A}{p}\right)=-1$, then we have $$S_p(A)+S_p\left(A^{-1}\right)=-2\,.\tag{1}$$

From the claim in this answer, we have $$-1=-\left(\frac{1}{p}\right)=\sum_{y\in\mathbb{F}_p}\,\left(\frac{y(y+A)}{p}\right)=\frac{1}{2}\,\sum_{x\in\mathbb{F}_p^{\times}}\,\left(\frac{x^2+A}{p}\right)+\frac{1}{2}\,\sum_{x\in\mathbb{F}_p^{\times}}\,\left(\frac{x^2+Au}{p}\right)\,,$$ where $u$ is an arbitrary quadratic nonresidue modulo $p$. Hence, $$S_p(A)+S_p(Au)=-2\,.\tag{2}$$

If $\left(\frac{A}{p}\right)=-1$, then (2) with $u:=A^{-1}$ yields $$S_p(A)+S_p(1)=-2\,.\tag{3}$$ Comparing (3) with (1), we have $S_p\left(A^{-1}\right)=S_p(1)$. However, this is true for all $A$ with $\left(\frac{A}{p}\right)=-1$, whence $S_p(A)=S_p(1)$. Consequently, $S_p(A)=-1$.

Now, if $\left(\frac{A}{p}\right)=+1$, then we know from the paragraph above that $S_p(Au)=-1$. Ergo, $S_p(A)=-1$ as well.

In general, let $p$ be an odd prime natural number and $a,b,c\in\mathbb{F}_p$. We have $$\sum_{x\in\mathbb{F}_p}\,\left(\frac{ax^2+bx+c}{p}\right)=\begin{cases} p\,\left(\frac{c}{p}\right)\,,&\mbox{if }a=0\mbox{ and }b=0\,,\\ 0\,,&\mbox{if }a=0\mbox{ and }b\neq0\,,\\ (p-1)\,\left(\frac{a}{p}\right)\,,&\mbox{if }a\neq 0\mbox{ and }b^2-4ac=0\,,\\ -\left(\frac{a}{p}\right)\,,&\mbox{if }a\neq 0\mbox{ and }b^2-4ac\neq 0\,.\\ \end{cases}$$

  • $\begingroup$ Another approach to the "In general" identity: Let $L = \sum\limits_{x \in \mathbb{F}_p} \left(\dfrac{ax^2+bx+c}{p}\right)$. Then, $L \in \left\{-p,-p+1,\ldots,p\right\}$. But recall that $\left(\dfrac{u}{p}\right) \equiv u^{\left(p-1\right)/2} \mod p$ for all integers $u$. Thus, $L \equiv \sum\limits_{x \in \mathbb{F}_p} \left(ax^2+bx+c\right)^{\left(p-1\right)/2} \mod p$. But $\left(ax^2+bx+c\right)^{\left(p-1\right)/2}$ is a polynomial of degree $\leq p-1$ in $x$, and its $x^{p-1}$-coefficient is $a^{\left(p-1\right)/2}$. Hence, ... $\endgroup$ – darij grinberg Aug 8 at 8:19
  • $\begingroup$ ... we know that $L$ is congruent (mod $p$) to the sum of all values of this polynomial over $\mathbb{F}_p$. However, it is known that such a sum is always congruent to minus the $x^{p-1}$-coefficient (since $\sum_{x \in \mathbb{F}_p} x^i = 0$ for all $i < p-1$ and since $\sum_{x \in \mathbb{F}_p} x^{p-1} = -1$). Thus, $L \equiv - a^{\left(p-1\right)/2} \equiv - \left(\dfrac{a}{p}\right) \mod p$. Combining this with $L \in \left\{-p,-p+1,\ldots,p\right\}$, we have thus narrowed down the value of $L$ to two or three options. But we also know the parity of $L$: If $b^2 - 4ac \neq 0$, then ... $\endgroup$ – darij grinberg Aug 8 at 8:22
  • $\begingroup$ ... (assuming $a \neq 0$) either none or two of the addends of the sum $L = \sum\limits_{x \in \mathbb{F}_p} \left(\dfrac{ax^2+bx+c}{p}\right)$ are zero (because the discriminant of a quadratic polynomial tells us whether it has exactly one root), and thus $L$ is odd (since all remaining addends in this sum are odd); otherwise, exactly one of these addends is zero, and thus $L$ is even. In either case, this uniquely determines the value of $L$ when $a \neq 0$. The case $a = 0$ is easy and can be handled separately. $\endgroup$ – darij grinberg Aug 8 at 8:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.